Live signing

1. Connect, then send {"type": "start", "lang": "rw", "token": "sk_live_…"}. The key goes in the message because browsers can't set headers on a WebSocket.
2. Wait for {"type": "ready"}, then send the audio as binary messages: 16 kHz, mono, signed 16-bit little-endian PCM, in chunks of 20 to 100 ms.
3. Words come back as final messages. Send each one to /v1/sign and play the signs one after another, in id order.
4. Send {"type": "stop"} to finish. The rest comes back, then {"type": "done"}.

import WebSocket from "ws"; // npm i ws

const ws = new WebSocket("wss://api.usesigna.org/v1/live");

ws.on("open", () => {
  ws.send(JSON.stringify({ type: "start", lang: "rw", token: process.env.SIGNA_KEY }));
});

ws.on("message", (data, isBinary) => {
  if (isBinary) return;
  const msg = JSON.parse(data.toString());
  if (msg.type === "ready") {
    // Start sending audio: 16 kHz, mono, 16-bit PCM, in 20-100 ms chunks.
    studioAudio.on("data", (pcm) => ws.send(pcm));
  }
  if (msg.type === "partial") showCaption(msg.text);
  if (msg.type === "final") signNext(msg.text); // POST it to /v1/sign, in order
  if (msg.type === "error") console.error(msg.message);
});

Words are approved while the speaker is still talking: a word that comes out the same in two readings in a row is sent straight away, so signing starts within a second or two even for someone who never pauses.

Most sources need converting. ffmpeg turns almost anything, a studio feed included, into the right format:

ffmpeg -i <your input> -f s16le -ac 1 -ar 16000 -

Read its output in chunks and send each chunk as one binary message.

• A session can run for up to 12 hours, or as long as your live minutes this month allow.
• If no audio arrives for 30 seconds the session closes, so keep sending through breaks (silence is fine).
• Time connected counts against your live minutes; signing the finals counts as characters, as usual.